Claude Code with a 1M token context window: when it helps
Some models run Claude Code with a million-token context. What that changes, when it is worth using, the costs to know about, and why a percentage means more.
Claude Code's context window is 200,000 tokens by default, and some models offer an extended window of a million. That is room for a large codebase, long logs and a long conversation at once. It is also easy to use badly.
When it helps
- Understanding a large, unfamiliar codebase where many files matter at once.
- Long investigations with big logs or traces.
- Refactors that touch many files and need to keep all of them in view.
When it does not
- Focused tasks on a few files. A smaller context is quicker and uses less.
- Sessions that grew large by accident, full of old tool output. Clearing is better than more room.
The costs
Every turn reads what is in context. More context means more tokens per turn, which uses your plan's limits faster, or costs more on an API key. Check the current pricing and plan details for the model you use, since long context is often priced differently.
Reading the percentage
The status line gets both used_percentage and context_window_size. On a million-token window, 20 per cent is 200,000 tokens, a full default window's worth. A status line that shows the percentage should make the size visible too, or at least the model, so you know which scale you are on. More on the number in seeing your context usage.
Load less, even with more room
A larger window does not change the habit that keeps sessions good: load what the task needs. Skills that keep detail in reference files, as in the references folder, help at any size.
Capsule Status Line shows the model with the context capsule, reads the window size from Claude Code rather than assuming 200,000, and changes colour at the same thresholds on either size.
Questions
How big is Claude Code's context window?
200,000 tokens by default. Models with extended context can use 1,000,000. The status line data reports the current size as context_window_size.
Is a bigger context always better?
No. It helps when a task needs a lot loaded at once. It uses more of your limits or budget per turn, and a focused session with less loaded is often faster and just as good.
Does 50% mean the same thing on both sizes?
No. Half of a million tokens is five times half of 200,000. The percentage tells you how close you are to the edge; the size tells you how much is being read each turn.